TCN-Transformer模型在鄂尔多斯盆地长8储层孔隙度预测精 准评价

打开文本图片集

中图分类号：P618 文献标识码：A 文章编号：2097-5465（2025）05-0048-09

Abstract：Porosityisacriticalparameterinreservoirevaluation.However，traditionalpredictionmethodssuferfromlimited accuracyduetotheirinabilitytoefectivelymodelcomplexnonlineargeologicalrelationships.Toimprovereservoirparameter prediction，thisstudyproposesanovelhybridmodelintegratingTemporalConvolutionalNetwork（TCN）andTransformer architectures.KeyinputloggingdatawereselectedbasedonPearsoncorelationcoefficents，andgeneticalgorihmwaseployed forhyperparameteroptimization.ThemodelwasappiedtotheChang8oillayergroupinthesouthwesternOrdos Basin，andits performance wascompared withstandaloneTransformer，CNN，andTCNmodels.ExperimentalresultsdemonstratethattheTCNTransformer modelachievessuperior performance，exhibiting lower mean absoluteerror（MAE）androtmeansquareeror （RMSE），alongwith a coefficient of determination（ R2 ）closer to1，indicating higherprediction accuracy.Furthermore，the modelshowsstronggeneralizationcapabilty，yieldingthelowestpredictionerrorsonanindependenttestsetwithoutadditional trainingorparameteruning.Thismethodprovidesahig-precisiontolforporosityprediction，oferingpracticalvalueforsevoir characterization and development planning.

KeyWords：porosityprediction；deep learning；TCN-Transformer model；loggingdataanalysis；geneticalgorithm

0 引言

孔隙度直接影响储层的储集能力和流体的流动性，这对油气储量和开采效率至关重要。（剩余8876字）